Automatic region selection method to enhance image-based steganography

Abstract

Image-based steganography is an essential procedure with several practical applications related to information security, user authentication, copyright protection, etc. However, most existing image-based steganographic techniques assume that the pixels that hide the data can be chosen freely, such as random pixel selection, without considering the contents of the input image. So, the “region of interest” such as human faces in the input image might have defected after data hiding even at a low inserting rate, and this will reduce the visual quality especially for the photos containing several human faces. With this view, we proposed a novel approach that combines human skin-color detection along with the LSB approach which can choose the embedding regions. The idea behind that is based on the fact that the Human Vision System HVS tends to focus its attention on selectively certain structures of the visual scene instead of the whole image. Practically, human skin-color is good evidence of the existence of human targets in images. To the best of our knowledge, this is the first attempt that employs skin detection in application to steganography which considers the contents of the input image and consequently can choose the embedding regions. Moreover, an enhanced RSA algorithm and Elliptic Curve Equation are used to offer a double level of security. In addition, the system embeds noise bits into the resulting stego-image to make the attacker’s task more confusing. Two datasets are used for testing and evaluation. The proposed scheme achieves minimum visual defects with double level of security.